Why AI in drug discovery has great potential and will be a significant advancement
AI in drug discovery holds enormous potential to speed up the time to market of new drugs. Historically, there has been a laborious trial-and-error process to identify compounds that will be successful, but now, an AI technique known as natural language processing is being used to predict the makeup of a new drug molecule1. This can be used because biological codes (proteins) are represented by a series of letters (amino acids) that are used by scientists to denote the overall drug1. Natural language algorithms (such as ChatGPT) can be applied in this instance to biological data to create a protein-language model that predicts the sequence of a new drug molecule1. This would shrink “the time required for the early stages of drug discovery from years to months”1. This is the first reason that AI in drug discovery will be a significant advancement, due to the efficiency and speed it will produce for new molecules in the future.
Not explicitly mentioned but insinuated from the increased efficiency and speed is the cost reduction achieved from AI use in drug discovery. These sorts of predictive algorithms can help differentiate which compounds are more likely to succeed3. This would limit the number of failed experiments, employee salaries, and overall research and development costs that could have previously worked for years to create a molecule that did not pass clinical trials. Essentially, AI in drug discovery would reduce the unnecessary resources spent on unsuccessful drug candidates and will help the bottom line of pharmaceutical companies pursuing development.
Protein based drugs, aka the ones that can be predicted with this sort of natural language algorithms, treat a variety of diseases ranging from cancers and immune disorders to infections and more4. Therefore, the hope is that AI can increase the effectiveness of existing drugs and also be used to create new molecules for diseases without current treatments, such as ALS1. The potential here is for AI to form entirely new classes or drugs that do not currently exist. This is called “de novo” design or synthesizing new molecules from scratch1. J&J has already used machine learning to help design an experimental cencer drug that is scheduled to start human testing next year2.
In terms of enhancing existing molecules, given a starting point, natural language models can “recommend tweaks to its amino acid sequence to improve its therapeutic benefit”1. A more relevant existing application of this was from a paper published in the National Academy of Sciences where protein-language models were used to expand the number of variants treated by a COVID-19 drug candidate1. The room for increasing efficacy of existing treatments plus the innovation and novelty that can result from AI proposals is another reason why AI in drug discovery poses significant advancement.
Another application of AI in drug discovery that significantly advances the field is the ability for increased precision and personalization. With real patient data, AI is analyzing information of individuals and therefore created the opportunity to tailor treatments to specific patient populations or even individuals3. J&J is hoping that by using the data it has collected smartly, it can help answer questions about the molecular traits of disease and how to target drugs to explicitly target those traits2. They have already been able to identify thousands of genetic variants that influence levels of certain blood proteins, about 80% which were not previously known2. Additionally, J&J hopes to leverage this AI in personalized medicine by spotting patterns that code for the gene-protein links to diseases. Compared to the previous speed of industry scientists scouring academic papers to identify these patterns, AI would be able to spot it in record time2.
Some of the risks that come with using natural language processing for drug discovery includes introducing unintended side effects, and if this does happen, taking the same precaution to ensure that clinical trials are not shortchanged until the full effects are known.
What is additionally spearheading this progress of AI in drug development is the explosion of biological data that is available3. For example, this article mentions one of J&J’s massive datasets purchased from a database called med.AI that includes real world patient data and clinical trial results. Because the amount of data available is increasing, data analysis techniques advancing, and algorithms improving, progress of AI in drug discovery is rampant. The result: the chief data science officer and global head of strategy and operations for J&J’s pharmaceutical research unit stated that “AI and data science are going to be the heart of how we are transforming and innovating”2.
References:
2. https://www.wsj.com/tech/biotech/johnson-johnson-hiring-data-scientists-ai-ccbf2c07
3. ChatGPT 3.5: “Why is AI in drug discovery going to be the next big thing”; 02DEC2023